VisualLens: Personalization through Task-Agnostic Visual History

Yue Liu (NUS) · Xin Dong (Harvard University) · Anuj Kumar (Facebook) · Seungwhan Moon (Facebook) · Yi Lu (Facebook) · Wang Bill Zhu (University of Southern California) · Deqing Fu (USC) · Kai Sun (Meta) · Zhaojiang Lin (The Hong Kong University of Science and Technology) · Kanika Narang (Facebook) · MUSTAFA CANIM (Facebook)
benchmarksgoogle-review-vhistory lengthshit@3multimodal large language modelsmultimodal recommendationpersonalizationrecommendation systemsrobustnesstask-agnostic insightsunseen content categoriesuser profile extractionvisual historyvisuallensyelp-v

Existing recommendation systems either rely on user interaction logs, such as online shopping history for shopping recommendations, or focus on text signals. However, item-based histories are not always accessible and generalizable for multimodal recommendation. We hypothesize that a user's visual history